Pages: [1]   Go Down

Author Topic: Why is Adobe's demosaicing algorithm so out-of-date?  (Read 3066 times)

ablankertz

  • Newbie
  • *
  • Offline Offline
  • Posts: 20
Why is Adobe's demosaicing algorithm so out-of-date?
« on: October 25, 2015, 08:55:37 am »

Amaze has now been around for five years or so, so Adobe's engineers have certainly had a chance to evaluate what's probably the best all around demosaicing algorithm and use it as the target to beat. So why can't they come close? I'm not even talking about detail here. Adobe's algorithm gives lousy noise compared to Amaze, and Amaze isn't even an algorithm optimized for low noise (e.g. LMMSE or IGV).
Logged

Rory

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 528
    • Recent images
Re: Why is Adobe's demosaicing algorithm so out-of-date?
« Reply #1 on: October 25, 2015, 10:02:07 am »

Amaze has now been around for five years or so, so Adobe's engineers have certainly had a chance to evaluate what's probably the best all around demosaicing algorithm and use it as the target to beat. So why can't they come close? I'm not even talking about detail here. Adobe's algorithm gives lousy noise compared to Amaze, and Amaze isn't even an algorithm optimized for low noise (e.g. LMMSE or IGV).

Can you show some examples with a sample raw file so we can see for ourselves as well.  I frequently hear how bad Adobe rendering is, but usually when I investigate, either the differences are too small for me to care or ACR/LR has not been used optimally.  So I want to see some evidence.  Many thanks in advance.
Logged
[url=http://www.flickr.com/photos/roryhi

Jimbo57

  • Full Member
  • ***
  • Offline Offline
  • Posts: 180
Re: Why is Adobe's demosaicing algorithm so out-of-date?
« Reply #2 on: October 25, 2015, 11:42:11 am »

Can you show some examples with a sample raw file so we can see for ourselves as well.  I frequently hear how bad Adobe rendering is, but usually when I investigate, either the differences are too small for me to care or ACR/LR has not been used optimally.  So I want to see some evidence.  Many thanks in advance.

Me too.

I tend to be one of the most critical (and self-critical) photographers around and I have never noticed any problem with this in Lightroom-processed photographs.
Logged

AlterEgo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1995
Re: Why is Adobe's demosaicing algorithm so out-of-date?
« Reply #3 on: October 25, 2015, 11:59:21 am »

Amaze has now been around for five years or so, so Adobe's engineers have certainly had a chance to evaluate what's probably the best all around demosaicing algorithm and use it as the target to beat.
to introduce something like a new demosaick, NR or sharpening they need to introduce a new "Process version" (just like Process 2003, 2010, 2012 now) to lock in the old/existing renderings for the users that used current demosaick and then Adobe converters are commercial products, the mere fact that some product out there has a better demosaick does not make a dent really... take a look at C1 for example ? they followed LR with DAM features, not vice versa ... that shows what actually works in the grand picture.
Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11250
    • Echophoto
Re: Why is Adobe's demosaicing algorithm so out-of-date?
« Reply #4 on: October 25, 2015, 12:48:08 pm »

Hi,

Check this: http://echophoto.dnsalias.net/ekr/Articles/A7rIIJourney/Demosaic/

Also this thread: http://forum.luminous-landscape.com/index.php?topic=104708.0

My take on the issue is that there is a tendency to non OLP filtered raw images. Perceived sharpness improves when the OLP filtering is lost so photographers love non AA (Anti Aliasing) filtered images. Unfortunately, with present pixel sizes we get a lot of aliasing when OLP filtering is dropped. Some algorithms are better in hiding aliasing artefacts than others. ACR and Lightroom are about the worst on non OLP filtered images.

Now, it could be argued that a properly designed camera would not alias, that would take a decent OLP-filter and small pixels, but market has chosen large pixels and no OLP filtering, so raw processors should try to cover up the design faults of the imaging system.

AMaZE is good doing that while ACR is notů

Best regards
Erik



Can you show some examples with a sample raw file so we can see for ourselves as well.  I frequently hear how bad Adobe rendering is, but usually when I investigate, either the differences are too small for me to care or ACR/LR has not been used optimally.  So I want to see some evidence.  Many thanks in advance.
Logged
Erik Kaffehr
 

AlterEgo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1995
Re: Why is Adobe's demosaicing algorithm so out-of-date?
« Reply #5 on: October 25, 2015, 05:40:13 pm »

ACR and Lightroom are about the worst on non OLP filtered images.
probably because when Process 2012 was prepped and that was before Jan 2012 there were few AA-less cameras (D800 release, for example = Feb 2012) with the intended mass audience for Adobe (not counting MF here)
Logged

Rory

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 528
    • Recent images
Re: Why is Adobe's demosaicing algorithm so out-of-date?
« Reply #6 on: October 25, 2015, 09:26:31 pm »

Hi,

Check this: http://echophoto.dnsalias.net/ekr/Articles/A7rIIJourney/Demosaic/

Also this thread: http://forum.luminous-landscape.com/index.php?topic=104708.0

My take on the issue is that there is a tendency to non OLP filtered raw images. Perceived sharpness improves when the OLP filtering is lost so photographers love non AA (Anti Aliasing) filtered images. Unfortunately, with present pixel sizes we get a lot of aliasing when OLP filtering is dropped. Some algorithms are better in hiding aliasing artefacts than others. ACR and Lightroom are about the worst on non OLP filtered images.

Now, it could be argued that a properly designed camera would not alias, that would take a decent OLP-filter and small pixels, but market has chosen large pixels and no OLP filtering, so raw processors should try to cover up the design faults of the imaging system.

AMaZE is good doing that while ACR is notů

Best regards
Erik

Okay, the difference is obvious at 300%, not so much at 100%, but yes, AMaZE looks pretty good.  Is there a hit on performance using the AMaZE algorithm?  Thanks for the example Erik!
Logged
[url=http://www.flickr.com/photos/roryhi
Pages: [1]   Go Up